Ana/typed mapped column

[acanizares]

NOTE: Additional changes are required before this can be merged.
Discussion: sqlalchemy#11093

Description

Summary
These are some efforts to get better typing of mapped_column. I believe the return type of mapped_column cannot be expressed correctly as it is now. It would be required to change the default value of nullable to get types correct.

Context
Right now, mapped_column returns MappedColumn[Any] so there is no guarantee that the type annotation matches the SQLA type engine passed to mapped_column (if any).
sqlalchemy#11093

When I first started the discussion I was hopeful that this could be fully solved with overloads. But now I think it is just not possible to type mapped_column correctly as long as the default value for nullable is True. Because then
there is no way to type something like mapped_column(Integer) correctly. mapped_column(Integer) can either be of type MappedColumn[int] or MappedColumn[Optional[int]], and the default should be the more specific type of those two, namely MappedColumn[int] since both of the following assignments should be valid:

x: Mapped[int]  = mapped_column(Integer)
y: Mapped[Optional[int]]  = mapped_column(Integer)

In this PR:

1. Modified type annotations in mapped_column to

• Return MappedColumn type matching the type engine, the nullable and the default parameters.

2. Made TypeEngine covariant. This was necessary for assignments such as

x: Mapped[Optional[int]] = mapped_column(Integer)

to be valid for the typechecker.

3. Added typing tests. This section gives a good overview of what this change is about.